A method for estimating user electricity consumption pattern profile based on optimal weighted optimization

Abstract

The present invention discloses a method for estimating user electricity usage profiles based on optimal weighted optimization, comprising the following steps: S1. For N users of the same type, each user's original electricity usage data is subjected to noise processing to obtain noisy user electricity usage data for each user; S2. The electricity usage profiles of the N users of the same type are estimated, and the expected value and mean square error (MSE) of the estimated results are determined; S3. The noisy user electricity usage data of each user is subjected to weighted average estimation according to time periods to obtain an estimated user electricity usage profile based on optimal weighted optimization. The present invention can effectively utilize noisy user electricity usage data and minimize the variance of the estimated value by calculating the optimal spatiotemporal weights, thereby obtaining the optimal user profile estimation result.

Description

Technical Field

[0001] The present invention relates to user electricity consumption pattern portrait estimation, and in particular to a user electricity consumption pattern portrait estimation method based on optimal weighted optimization. Background Art

[0002] Smart meters are widely deployed in residential areas, collecting vast amounts of user electricity usage data. This data effectively facilitates economic dispatch of distribution networks and efficient demand-side management, improving consumer-oriented decision-making and data analysis. Customized services based on this data improve economic efficiency, but the massive collection of user data has raised public concerns about privacy leaks. Consequently, the power industry is employing various methods and mechanisms to protect user privacy, such as federated learning and multi-party computation algorithms. While these privacy-preserving mechanisms do not compromise the accuracy of load profile analysis, they do increase communication, hardware, and computing costs. Furthermore, most of these methods are highly customized, requiring readjustment or even reconfiguration when switching between application scenarios.

[0003] Therefore, the power industry needs to introduce new privacy protection mechanisms. Adding noise to raw data is a classic privacy protection mechanism that doesn't increase hardware and communication costs, but it often suffers from insufficient data accuracy. How to maximize the use of noisy data to estimate user electricity usage patterns and minimize the error in these estimates is an important research question. Summary of the Invention

[0004] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a user electricity consumption pattern portrait estimation method based on optimal weighted optimization, which can effectively utilize noisy user electricity consumption data and minimize the variance of the estimated value by calculating the optimal spatiotemporal weights, thereby obtaining the optimal user portrait estimation result.

[0005] The object of the present invention is achieved through the following technical solution: a method for estimating a user's electricity consumption pattern profile based on optimal weighted optimization, comprising the following steps:

[0006] S1. For N users of the same type, perform noise processing on the original electricity consumption data of each user to obtain the noisy user electricity consumption data of each user;

[0007] S2. Estimate the electricity usage profiles of N users of the same type and determine the expected value and mean square error of the estimated results.

[0008] S3. Perform weighted average estimation on the noisy user electricity consumption data of each user according to time periods to obtain the user electricity consumption pattern portrait estimation result based on optimal weighted optimization.

[0009] The beneficial effect of the present invention is that the present invention can effectively utilize noisy user electricity consumption data, minimize the variance of the estimated value by calculating the optimal spatiotemporal weights, and thus obtain the optimal user portrait estimation result. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 is a flow chart of the method of the present invention;

[0011] Figure 2 Schematic diagram of the distribution of estimation errors in the embodiment;

[0012] Figure 3 Schematic diagram of the probability of occurrence of large deviations in the embodiment. DETAILED DESCRIPTION

[0013] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the following.

[0014] like Figure 1 As shown, a method for estimating a user's electricity consumption pattern profile based on optimal weighted optimization includes the following steps:

[0015] S1. For N users of the same type, perform noise processing on the original electricity consumption data of each user to obtain the noisy user electricity consumption data of each user;

[0016] Assume that the original data of a user's electricity consumption is d, where d=(d1,d2,d3,…,d T ) is the electricity consumption data of the user at different time points, and T is the length of time the user uses electricity. After Gaussian noise addition, Where η is a T-dimensional random variable that obeys a Gaussian distribution. The elements of each dimension are independent and obey an unbiased Gaussian distribution. Its distribution function satisfies:

[0017]

[0018] Where σ is the standard deviation of the Gaussian distribution, π is the pi constant, and the domain of the variable x is (-∞, +∞);

[0019] S2. Estimate the electricity usage profiles of N users of the same type and determine the expected value and mean square error of the estimated results.

[0020] A household profile refers to the typical electricity consumption pattern of a type of user, which is generally considered to be the average value of the electricity consumption curve of this type of user.

[0021] Consider a decision maker who needs to estimate the electricity consumption profile of users of type p. Consider that the decision maker has K+1 types of data with different Gaussian noise added, and these types are 0, 1, 2, ..., K. The data of type k is added with the distribution Gaussian noise. The decision maker has a total of N data of any type k. k The total amount of data he has Data demanders seek to obtain the most accurate estimate of user profiles based on this data. First, consider the simplest and most common method to obtain user profile estimates, namely the sample average:

[0022]

[0023] in, represents a noisy user electricity consumption data set with type k Gaussian noise added; Represents the noisy user electricity consumption data of the i-th user:

[0024]

[0025] Among them, η (i) represents the Gaussian noise injected into the i-th user, d (i) represents the electricity consumption data of the i-th user;

[0026] Then, the sample mean estimate is The expectation and mean square error (MeanSquaredError) satisfy,

[0027]

[0028]

[0029] Where, μ t is the value of the precise user portrait dimension t of type p user, Var(d t )express The variance of represents the electricity consumption of user i at time t.

[0030] This result directly shows that the sample mean estimate is unbiased and The estimated mean square error of decreases as the total amount of data N increases. Further, we can observe that the right-hand side term in the brackets in the equation is the weighted mean variance, where the weight α is k It is proportional to the amount of data corresponding to the noise type k.

[0031] However, the sample mean estimate does not necessarily yield the minimum mean squared error with the given data. Let's use a simple example to illustrate this: Consider two noisy data points, A and B, with variances of 1 and 10, respectively. When we estimate using only data point A, the estimated mean squared error is 1. However, after adding data point B, the estimated mean squared error becomes 1 / 2 × (12 × 1 + 12 × 10) = 2.75, which is worse than before.

[0032] More specifically, when the variance σ of a new noise data 2 When the following conditions are met, the mean square error will increase after introducing this data into the estimation:

[0033]

[0034] We can see that the value on the right is twice the weighted mean square error of the data. This shows that data with large noise (approximately twice the mean noise) has a negative impact on the estimation. To improve this estimation method, an optimal weighted average method is proposed to minimize the estimated mean square error.

[0035] S3. Perform weighted average estimation on the noisy user electricity consumption data of each user according to time periods to obtain the user electricity consumption pattern portrait estimation result based on optimal weighted optimization.

[0036] Intuitively, less noisy data contributes more to the estimate than noisy data. Therefore, simply averaging the different data is not good enough. Instead, giving them different weights and taking a weighted average can produce better estimates. This motivates us to perform the following weighted average estimation for different time periods t:

[0037]

[0038] where w t,k It is expressed as the weight factor designed for data with k-type noise at time t. For all weights w t,k ≥0, Obviously it is an unbiased estimate. The mean square error satisfies:

[0039]

[0040] Now we can choose the optimal weight w t,k To minimize the mean square error

[0041] make For variable w t,k Take the derivative and set it to 0 to get the optimal weight w t,k The value of , that is, the optimal weight, we can prove that the optimal weight that satisfies the minimum estimated mean square error is satisfy:

[0042]

[0043] In the embodiment of the present application, the performance evaluation of the optimal weighted optimization algorithm can also be performed.

[0044] For the optimal weights, the minimum mean square error satisfies

[0045]

[0046] The optimal weight is only related to Var(d t ) is related to the variance of the injected noise. The variance of the data with large noise is Leading to a smaller weight w t,k . And they are related to N k Furthermore, we found that by designing the optimal weights, we optimized the variance to a weighted geometric mean square error, which is theoretically smaller than the weighted mean square error obtained by averaging samples.

[0047] In the embodiments of the present application, the performance improvement of the optimal weighted estimation was verified. During the simulation process, the user electricity consumption dataset was used to conduct user portrait clustering research. We performed multiple random sampling and noise injection of user data, and compared the estimation deviation of the cluster center point by the optimal weighted estimation method and the sample average estimation method after adding noise.

[0048] Figure 2 shows the improvement of optimal weighted clustering over sample average clustering, specifically, Figure 3 The distribution of estimation errors is shown in Figure 2. Optimal weighted clustering can reduce the estimation error by 10.2%. In addition, for the case of large estimation deviations, Figure 2 It is shown that optimal weighted clustering can significantly reduce the probability of large estimation deviations, which proves the effectiveness of the optimal weighting method.

[0049] Although exemplary embodiments of the present invention have been described for illustrative purposes, it will be understood by those skilled in the art that various modifications, additions, and substitutions may be made in form and detail without departing from the scope and spirit of the invention disclosed in the appended claims, and all such changes shall fall within the scope of protection of the appended claims, and the various steps in the method claimed in the present invention may be combined in any combination. Therefore, the description of the embodiments disclosed in the present invention is not intended to limit the scope of the invention, but is used to describe the invention. Accordingly, the scope of the present invention is not limited by the above embodiments, but is defined by the claims or their equivalents.

Claims

1. A method for estimating a user's electricity consumption pattern profile based on optimal weighted optimization, characterized by: The following steps are involved: S1. For N users of the same type, perform noise processing on the original electricity consumption data of each user to obtain the noisy user electricity consumption data of each user; S2. Estimate the electricity usage profiles of N users of the same type and determine the expected value and mean square error of the estimated results. The step S2 comprises: S201. Assume that K+1 types of Gaussian noise are added in the noise addition process, and the types are 0, 1, 2, ..., K, where the Gaussian noise distribution of type k is k=0,1,2,…,K, the number of Gaussian noises of type k added is N k ,but S202. For N users of the current type, estimate the power usage profile. The user power usage profile is the average value of the user power usage data obtained after Gaussian noise addition, that is, the average estimated value of the sample, denoted as: in, represents a noisy user electricity consumption data set with type k Gaussian noise added; Represents the noisy user electricity consumption data of the i-th user: Among them, η (i) represents the Gaussian noise injected into the i-th user, d (i) represents the electricity consumption data of the i-th user; S203. Let the average estimated value of the samples obtained in step S202 be but The expectation and mean square error satisfy: Where, μ t is the value of the precise user portrait dimension t of the current type of user, Var(d t )express The variance of represents the electricity consumption of user i at time t; S3. Perform weighted average estimation on the noisy user electricity consumption data of each user according to time period to obtain the user electricity consumption pattern profile estimation result based on optimal weighted optimization; The step S3 comprises: S301. Perform the following weighted average estimation for different time periods t: where w t,k It is expressed as the weight factor designed for data with k-type noise at time t, for all weights w t,k ≥0, It is an unbiased estimate, and the mean square error satisfies: Determine the optimal weight w t,k To minimize the mean square error make For variable w t,k Take the derivative and set it to 0 to get the optimal weight w t,k The value of , that is, the optimal weight, minimizes the optimal weight of the estimated mean square error satisfy: Based on the optimal weight w t,k , user electricity usage pattern profile estimation Satisfies the following formula:

2. The method for estimating a user power consumption pattern profile based on optimal weighted optimization according to claim 1, characterized in that: The step S1 comprises: S101. Assume that the original electricity consumption data of any user is d, where d=(d1, d2, d3, ..., d T ), d1, d2, d3, …, d T represents the electricity consumption information corresponding to time 1, 2, ..., T, where T represents the time length of the user's original electricity consumption data d; After Gaussian noise is added to the user's original electricity consumption data, the noisy user electricity consumption data is obtained. Where η is a T-dimensional random variable that obeys a Gaussian distribution. The elements of each dimension are independent and obey an unbiased Gaussian distribution. Its distribution function satisfies: Where σ is the standard deviation of the Gaussian distribution, π is the pi constant, and the domain of the variable x is (-∞, +∞); S102. For N users of the same type, execute step S101 on their original electricity usage data to obtain N noisy user electricity usage data.

Images